Procedure

The experiment was conducted by using an online questionnaire, whereby upon the start of the survey participants were randomly assigned to one of the eight experimental conditions. At the start of the survey participants were presented with a brief introduction to the study and were informed of their right to leave the questionnaire at any given point and on the treatment of data. The survey started after participants agreed to participate.

In the first part of the online questionnaire, participants were confronted with four brands together with a short description and were asked to choose the brand that either fits them the most or the least (dependent on the condition). Followed by this, participants were presented with the screenshots of the Instagram accounts of four Instagrammers together with a short description of these individuals. After viewing the Instagram accounts and the

31 descriptions of each of these influencers, participants were also here instructed to choose one of the influencers with who they either identify the most or the least.

Subsequently, once participants had chosen an influencer and a brand, the two were paired together and either displayed a brand-endorser match or a brand-endorser mismatch. Participants were here presented with an Instagram post in which the influencer is endorsing the brand, together with a short text stating whether this concerns a match/mismatch and a brief explanation why. See figure 6 for an example of such an Instagram post with a brand- endorser combination and an explanation for the match or mismatch.

Figure 6. Example of an Instagram post with a brand-endorser mismatch (based on the choice for influencer Nikkie Tutorials and brand Patagonia)

32 The matches and mismatches were thus determined beforehand in order to fill in all the

experimental conditions. This means that an actual match between the endorser and brand could be displayed as a mismatch and vice versa. In the text, however, there was an explanation on why the brand and endorser are a fit/misfit in order for the participants to perceive the pair as an actual match/mismatch.

Followed by this, participants were asked to answer a series of questions regarding ad attitude and online engagement, followed by the manipulation check which assessed the perceived brand-endorser congruency. Consequently, participants were presented with questions assessing brand attitude, purchase intention and the perceived brand-consumer congruency, followed by items regarding source credibility and the perceived endorser- consumer congruency. Finally, the survey ended with a couple of demographic questions. Since it was important for participants to enter the study as objectively as possible, they were presented with the research objective after the questionnaire was completed. After that, they were thanked for their participation and could afterward close the online questionnaire. See Appendix D for the complete questionnaire.

3.5 Measures

The questionnaire contained different constructs in accordance with the research question and the formulated hypotheses. The used measures were divided into subjects regarding the constructs of source credibility, ad attitude, brand attitude, purchase intention, online engagement, and manipulation checks, and were composed by using several measurements scales from existing literature which were empirically validated and have proven their reliability in prior studies.

33

3.5.1 Dependent variables

Ad attitude

The dependent variable of ad attitude was measured by a seven-point semantic differential scale consisting of nine-items adapted from Greaff (1996). In this scale, a statement (‘’In my opinion the Instagram is…’’) was given for all items, which participants had to complete by judging their attitudes toward the Instagram post. For instance, items ranged from ‘’boring’’ to ‘’interesting’’, ‘’unsurprising’’ to ‘’surprising’’, ‘’unexciting’’ to ‘’intriguing’’, and ‘’favorable’’ to ‘’unfavorable’’ (Graeff, 1996). See Appendix D for all the measure items. Reliability analysis showed a high reliability (α = .87) for this scale.

Brand attitude

The second dependent variable, brand attitude, was measured by implementing a seven-point semantic differential scale consisting of eight items. Five of these items were developed by Greaff (1996) and three items were created based on the findings of the pre-test. Also here participants were given a statement (‘’In my opinion the brand is…’’) which they had to complete by judging their attitude regarding the shown brand. For instance, items ranged from ‘’unappealing’’ to ‘’appealing’’, ‘’unlikeable’’ to ‘’ likeable’’, and ‘’pleasant’’ to

‘’unpleasant’’ (Graeff, 1996). The three items created based on the pre-test ranged from ‘’expensive’’ to ‘’cheap’’, ‘’of low quality’’ to ‘’of high quality’’, and ‘’high-end’’ to ‘’high- street’’. Alpha reliability for this scale was α = .83.

Purchase intention

The third dependent variable of purchase intention was measured by a seven-point Likert scale consisting of four items adapted by Baker and Churchill (1977). This scale gave one statement per item which participants rated by agreement (1 = “strongly disagree”, 7 = “strongly agree”). The statements which were used for measuring the construct are ‘’I would

34 like to try the brand’’, ‘’I would buy the product if I happened to see the brand’’, ‘’I would actively seek out this brand to purchase it’’, and ‘’I would be willing to buy products from this brand’’. Reliability was also proven to be high for this scale (α = .91).

Online engagement

The final dependent variable, online engagement, was implemented by a seven-point Likert scale consisting of four items. This measurement scale was not adapted from earlier studies and was composed based on the interactions that exist on social media, in particular, the social media platform Instagram. Also here the scale gave one statement per item which participants rated by agreement (1 = “strongly disagree”, 7 = “strongly agree”). The statements which were used for measuring online engagement are ‘’I would share this Instagram post’’, ‘’I would comment on this Instagram post’’, ‘’I would like this Instagram post’’, and ‘’I would follow this Instagram account’’. Reliability analysis showed a high reliability (α = .78) for this scale.

3.5.2 Mediators

Source credibility

The mediator of source credibility consisted of the dimensions of attractiveness, expertise, and trustworthiness and was measured by implementing a five-point semantic differential scale, consisting of seventeen items. Fifteen of these items were taken from Ohanian (1990) and two items were created based on the results of the pre-test. In this scale, a statement (‘’I consider the Instagrammer as…”) was given for all the items, which participants had to complete by judging their perceived source credibility. For instance, items ranged from ‘’plain” to ‘’elegant’’, ‘’unknowledgeable’’ to ‘’knowledgeable’’, and ‘’insincere’’ to ‘’sincere’’ (Ohanian, 1990). The two items created based on the pre-test ranged from

35 high reliability for all three dimensions of source credibility (attractiveness α = .87; expertise α = .93; trustworthiness α = .88).

3.5.3 Manipulation checks

To ensure that participants understood the manipulations in the questionnaire, manipulations checks for the brand-endorser -, brand-consumer -, and endorser-consumer (in)congruent combinations were implemented. These manipulation checks were shown after exposure to the stimulus material.

Brand-endorser congruency

The first manipulation check question asked participants for their perception of the match between the brand and the endorser, and consisted of the following single statement: ‘’There is a good match between the brand and the Instagrammer’’. This statement was measured on a seven-point Likert scale which participants rated by agreement (1 = “strongly disagree”, 7 = “strongly agree”).

In order to determine if the congruent and incongruent conditions were perceived correctly and to find out whether the results for the manipulated variables differed

significantly, an independent sample t-test was performed. The criterion used to classify a pure match is everything above the score of four on the seven-point Likert scale, whereas everything under the score of four was considered as a pure mismatch. Although the results for the brand-endorser congruency showed a significant difference between the congruent condition (M = 4.05, SD = 1.73) and the incongruent condition (M = 2.92, SD = 1.85); t (204) = 4.51, p < .001), the brand-endorser combination in the congruent condition was not

perceived as a pure match as the mean score is slightly higher than the neutral score of four. Based on these findings it can thus be concluded that participants who were presented with a brand-endorser match perceived this combination as slightly congruent whilst the brand-

36 endorser combination in the incongruent condition was rated as a clear mismatch by

participants (see Table 4).

Brand-consumer congruency

The second manipulation check measured the perceived congruence between the brand and the consumer. Also for this question a single statement (‘’There is a good match between the brand and myself’’) was used for which participants had to indicate to what extent they agreed with the statement (1 = “strongly disagree”, 7 = “strongly agree”).

Also for this manipulation an independent sample t-test was conducted. The results show that the manipulation of the brand-consumer combination was successful and reveal significant differences as well. Participants in the congruent brand-consumer condition scored significantly higher on the manipulation check question concerning the match between the chosen brand and the participant (M = 5.14, SD = 1.39), than participants in the incongruent brand-consumer condition (M = 2.68, SD = 1.45); t (204) = 12.42, p < .001). This leads to the conclusion that the manipulation with regard to the match between the brand and consumer worked and that participants mostly perceived the congruent combination as a pure fit and the incongruent combination as a clear misfit.

Endorser-consumer congruency

The final manipulation check was implemented in order to measure the perceived match between the endorser and the consumer. This manipulation check consisted of three

statements which were also measured on a seven-point Likert scale (1 = “strongly disagree”, 7 = “strongly agree”). The statements used for this question are: ‘’The Instagrammer is similar to me’’, ‘’I can relate to the Instagrammer’’, and ‘’I can identify with the Instagrammer’’. Reliability analysis showed a high reliability (α = .95) for this scale.

37 Here the t-test revealed significant results as well between the congruent (M = 3.97, SD = 1.43) and incongruent endorser-consumer groups (M = 2.03, SD = 1.11); t (204) = 10.90, p < .001). However, just like the first congruency type, the endorser-consumer

combination in the congruent condition was not perceived as a pure match as the mean score is under the criterion score of four. Based on these findings it can be concluded that

participants in the congruent condition could identify more with the chosen endorser in comparison to participants in the incongruent condition and that participants mostly perceived the congruent combination as a moderate fit whereas the incongruent combination was rated as a pure mismatch.

Table 4

Manipulation check for congruency combinations (n = 206)

Match Mismatch Congruency type M (SD) M (SD) t p Brand-endorser congruency 4.05 (1.73) 2.92 (1.85) 4.51 0.00** Brand-consumer congruency 5.14 (1.39) 2.68 (1.45) 12.42 0.00** Endorser-consumer congruency 3.97 (1.43) 2.03 (1.11) 10.90 0.00** * p < .05, ** p < .01

Since the dataset still consists of participants who answered the manipulation questions wrongly, a second analysis was conducted without their responses. First the data for each participant who indicated a 3 or below on the seven-point Likert scale for the congruent conditions or a 5 or above for the incongruent conditions was removed from the dataset. Consequently, a median split technique (brand-endorser match median = 4; brand-consumer match median = 4; endorser-consumer median = 3) was used to split the dataset into two groups (congruent and incongruent). The independent samples t-test for this new dataset indicated for all the manipulation checks a significant difference and showed pure matches and mismatches for all three congruency types (see Table 5). Nevertheless, since no strong

38 differences have been found between the original dataset and this dataset in terms of results and since more than half of the original dataset would be discarded, which produces too little data (n = 99), the complete dataset (n = 206) is used for further analyses. However, it must be taken into account that not all participants answered the manipulation checks correctly.

Table 5

Manipulation check for congruency combinations after removal incorrect data (n = 99)

Match Mismatch Congruency type M (SD) M (SD) t p Brand-endorser congruency 5.08 (0.97) 1.84 (0.85) 17.84 0.00** Brand-consumer congruency 5.23 (0.99) 1.92 (0.80) 17.40 0.00** Endorser-consumer congruency 4.54 (0.82) 1.51 (0.56) 21.46 0.00** * p < .05, ** p < .01

39

4. Results